Environmental Policy
Meta analyse
Review articles
Review
Author: has the freedom to choose the article to write about
Concern: less objective than systematic review
Systematic review
Very clearly (methodological) write what you did during the search process of an
article
Gather all possible literature about the subject write how you worked put all
the information in research tables + mention why you didn’t take certain articles in
account
(Nauwkeurig alle literatuur verzamelen over onderwerp beschrijven hoe te werk
werd gegaan alle informatie in onderzoekstabellen zetten + zeggen waarom
bepaalde artikels niet meegenomen werden)
Rigorous review of specific clinical question
‘Systematic’ literature search
Explicit regarding information provided
Steps
Clearly formulated question
Identify literature (is there enough literature?)
Select studies that are valid
Describe quality of selected studies
May use forest plot to summarize studies
Identify potential bias with funnel plot (ex. publication bias: study gets
negative results doesn’t get published as often keep in mind that the
true results can differ because of this)
Meta-analysis
Gives a clearer picture (multiple studies better view on results)
Systematic review
Quantitative pooling of data
Statistical technique for combining individual effects of a number of studies addressing the
same question to produce a summary effect
, Studies with more accurate prediction (smaller CI) more weight in end result not an
average because some studies carry more weight
Weighting factor: 1/(standard error)2
Systematic review: flowchart
Justify why you used or didn’t use certain studies
Flowchart can be used
Use of search terms: ‘myocardial infarction’ and ‘trigger’
538 studies found
Only 17 studies used
Reasons? review article, double published paper, not a specific trigger,
not the right study by reading the abstract
New search to find other articles with new keywords: ‘myocardial infarction’ and
‘case-crossover’ or ‘time-series’
20 studies found
Total: 37 studies used in analysis
What is a meta-analysis?
Focus on mean effects and differences between studies
Determines effect on individuals
Quantification of individual responses as a standard deviation
Subject characteristics are known (age, gender)
Problem: quality of the studies is important
Forest plot
, Bottom one: mean
Heterogeneity
Clinical heterogeneity
Clinical differences in studies with respect to patients, interventions and outcomes
Methodological heterogeneity
Other methods used
Statistical heterogeneity
Individual trials may seem to measure same outcome, but have results that are not
consistent
test for heterogeneity:
‘null hypothesis’ – difference due to chance
p value
Lots of heterogeneity: hard to summarize studies
Examples of statistical heterogeneity:
2 studies: verbetering van bv medicatie
2 studies: geen verbetering
Heeft geen zin om meta analyse van te doen
Kijken of bv populatie verschilt of andere verklaringen waarom sommige studies wel
verbetering vinden en andere niet
Extreem: geen overlap tussen studies
, Minder sterke heterogeniteit alle studies gaan in dezelfde richting
Maar toch: beter om alle individuele studies te bekijken en te poolen om te zien
waar het verschil aan ligt
Think about these things:
Was it a good idea to combine the studies?
Is there too much clinical heterogeneity to make sense?
Do the forest plots look consistent?
Do statistical tests suggest that heterogeneity is a problem?
Estimating a common effect
Pooled results: provide a more precise estimate
Type of weighted average
Sample size: important when calculating weight
Event rate
Precision: 95% CI
Test for heterogeneity p > 0.2 not significant acceptable
, Sensitivity analysis
For sources of heterogeneity
To see how many studies it takes to change the result
Exclude studies one by one calculate estimate again notice which study modifies the
estimate the most
Publication bias
Positive results
More likely to be published
Rapidly published
In English
Cited
Literature may be selective and biased
Biased interpretation of trial results
Solution
Since 2005: clinical trial registration mandatory
Requires specific information about patient population, intervention and outcomes
Eerst informatie registreren voor het onderzoek uitgevoerd wordt zo kunnen
resultaten en bijgevolg publicaties niet achtergehouden worden die een ongewenst
resultaat uitkomen
Funnel plots
Detect bias in studies
Results of the studies are plotted against the sample size
Odds ratio: estimate
Vormt vorm als er geen publicatie bias is
Onderste bollen: kleine studies meer variatie
Meta analyse
Review articles
Review
Author: has the freedom to choose the article to write about
Concern: less objective than systematic review
Systematic review
Very clearly (methodological) write what you did during the search process of an
article
Gather all possible literature about the subject write how you worked put all
the information in research tables + mention why you didn’t take certain articles in
account
(Nauwkeurig alle literatuur verzamelen over onderwerp beschrijven hoe te werk
werd gegaan alle informatie in onderzoekstabellen zetten + zeggen waarom
bepaalde artikels niet meegenomen werden)
Rigorous review of specific clinical question
‘Systematic’ literature search
Explicit regarding information provided
Steps
Clearly formulated question
Identify literature (is there enough literature?)
Select studies that are valid
Describe quality of selected studies
May use forest plot to summarize studies
Identify potential bias with funnel plot (ex. publication bias: study gets
negative results doesn’t get published as often keep in mind that the
true results can differ because of this)
Meta-analysis
Gives a clearer picture (multiple studies better view on results)
Systematic review
Quantitative pooling of data
Statistical technique for combining individual effects of a number of studies addressing the
same question to produce a summary effect
, Studies with more accurate prediction (smaller CI) more weight in end result not an
average because some studies carry more weight
Weighting factor: 1/(standard error)2
Systematic review: flowchart
Justify why you used or didn’t use certain studies
Flowchart can be used
Use of search terms: ‘myocardial infarction’ and ‘trigger’
538 studies found
Only 17 studies used
Reasons? review article, double published paper, not a specific trigger,
not the right study by reading the abstract
New search to find other articles with new keywords: ‘myocardial infarction’ and
‘case-crossover’ or ‘time-series’
20 studies found
Total: 37 studies used in analysis
What is a meta-analysis?
Focus on mean effects and differences between studies
Determines effect on individuals
Quantification of individual responses as a standard deviation
Subject characteristics are known (age, gender)
Problem: quality of the studies is important
Forest plot
, Bottom one: mean
Heterogeneity
Clinical heterogeneity
Clinical differences in studies with respect to patients, interventions and outcomes
Methodological heterogeneity
Other methods used
Statistical heterogeneity
Individual trials may seem to measure same outcome, but have results that are not
consistent
test for heterogeneity:
‘null hypothesis’ – difference due to chance
p value
Lots of heterogeneity: hard to summarize studies
Examples of statistical heterogeneity:
2 studies: verbetering van bv medicatie
2 studies: geen verbetering
Heeft geen zin om meta analyse van te doen
Kijken of bv populatie verschilt of andere verklaringen waarom sommige studies wel
verbetering vinden en andere niet
Extreem: geen overlap tussen studies
, Minder sterke heterogeniteit alle studies gaan in dezelfde richting
Maar toch: beter om alle individuele studies te bekijken en te poolen om te zien
waar het verschil aan ligt
Think about these things:
Was it a good idea to combine the studies?
Is there too much clinical heterogeneity to make sense?
Do the forest plots look consistent?
Do statistical tests suggest that heterogeneity is a problem?
Estimating a common effect
Pooled results: provide a more precise estimate
Type of weighted average
Sample size: important when calculating weight
Event rate
Precision: 95% CI
Test for heterogeneity p > 0.2 not significant acceptable
, Sensitivity analysis
For sources of heterogeneity
To see how many studies it takes to change the result
Exclude studies one by one calculate estimate again notice which study modifies the
estimate the most
Publication bias
Positive results
More likely to be published
Rapidly published
In English
Cited
Literature may be selective and biased
Biased interpretation of trial results
Solution
Since 2005: clinical trial registration mandatory
Requires specific information about patient population, intervention and outcomes
Eerst informatie registreren voor het onderzoek uitgevoerd wordt zo kunnen
resultaten en bijgevolg publicaties niet achtergehouden worden die een ongewenst
resultaat uitkomen
Funnel plots
Detect bias in studies
Results of the studies are plotted against the sample size
Odds ratio: estimate
Vormt vorm als er geen publicatie bias is
Onderste bollen: kleine studies meer variatie
